Gemini API Skill for RunAPI

Configure Gemini or OpenAI-compatible clients to use Gemini models on RunAPI.

Model Reference · Skill Repo · All Models

Call the Gemini API through RunAPI with either the OpenAI SDK or Gemini contents clients -- point any OpenAI-compatible client at https://runapi.ai/v1, send any Gemini 2.5 or 3 model (see Supported models below), or call Gemini contents paths when an existing client expects generateContent / streamGenerateContent, and pay through one RunAPI balance. This skill teaches Claude Code, Codex, Gemini CLI, Cursor, and 50+ agents how to wire a Gemini API client up against RunAPI.

The canonical agent file is skills/gemini/SKILL.md.

Install the skill

npx skills add runapi-ai/gemini -g

Or paste this prompt to your AI agent:

Install the gemini skill for me:
1. Clone https://github.com/runapi-ai/gemini
2. Copy the skills/gemini/ directory into your
user-level skills directory (e.g. ~/.claude/skills/
for Claude Code, ~/.codex/skills/ for Codex).
3. Verify that SKILL.md is present.
4. Confirm the install path when done.

Use the Gemini API on RunAPI

The Gemini API on RunAPI exposes two request styles:

1. OpenAI-compatible — POST /v1/chat/completions. Drop-in for any existing OpenAI SDK client.
2. Gemini contents — POST /v1beta/models/<model>:generateContent or :streamGenerateContent. Drop-in for clients that send Gemini contents requests.

fromopenaiimportOpenAIclient=OpenAI(
api_key="YOUR_RUNAPI_TOKEN",
base_url="https://runapi.ai/v1",
)
response=client.chat.completions.create(
model="gemini-2.5-flash",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)

importOpenAIfrom"openai";constclient=newOpenAI({apiKey: "YOUR_RUNAPI_TOKEN",baseURL: "https://runapi.ai/v1",});constresponse=awaitclient.chat.completions.create({model: "gemini-2.5-flash",messages: [{role: "user",content: "Hello!"}],});console.log(response.choices[0].message.content);

curl -X POST "https://runapi.ai/v1/chat/completions" \
-H "x-api-key: YOUR_RUNAPI_TOKEN" \
-H "Content-Type: application/json" \
-d '{ "model": "gemini-2.5-flash", "messages": [{"role": "user", "content": "Hello!"}] }'

Get a RunAPI API Key at https://runapi.ai/api_keys.

For gemini-3.7-flash, gemini-3.6-flash, gemini-3.5-flash, and gemini-3-flash-preview, RunAPI uses the native Gemini streamGenerateContent route. For other callable Gemini models, RunAPI accepts Gemini contents requests and bridges them to the OpenAI-compatible chat request format. For new app code, prefer the OpenAI-compatible setup.

Connect Gemini CLI itself

export GOOGLE_API_KEY=YOUR_RUNAPI_TOKEN
export GOOGLE_GENAI_BASE_URL=https://runapi.ai
gemini

Supported Gemini models

Model ID	Capabilities
gemini-3.7-flash	Streaming contents and OpenAI-compatible chat; native functionDeclarations, the empty-object googleSearch tool, and thinkingConfig are protocol-specific
gemini-3.6-flash	Streaming contents and OpenAI-compatible chat, multimodal, function calling
gemini-3.5-flash	Streaming contents requests, multimodal, function calling, thoughts
gemini-3.1-pro-preview	Chat, multimodal, structured output, reasoning effort
gemini-3-flash-preview	Chat, multimodal, function calling, structured output, reasoning effort
gemini-2.5-pro	Chat, multimodal, Google Search, structured output, reasoning effort
gemini-2.5-flash	Chat, multimodal, Google Search, structured output, thoughts

gemini-flash-latest is an alias for gemini-3-flash-preview.

Routing

• Gemini API on RunAPI: https://runapi.ai/models/gemini
• Gemini API docs: https://runapi.ai/docs/api/gemini/generate-content
• Provider page: https://runapi.ai/providers/google
• Browse the full RunAPI catalog: https://runapi.ai/models
• Skill repository: https://github.com/runapi-ai/gemini

Agent rules

• Keep API keys in GOOGLE_API_KEY / OPENAI_API_KEY (or your secret manager); never inline them in commits or shell history.
• Use the OpenAI-compatible endpoint for new app code. Use Gemini contents paths when an existing client already sends contents requests.
• Stream long responses (stream: true) so the agent can release the terminal/IO loop early.
• For gemini-3.7-flash, use native search grounding only with the googleSearch tool and an empty object payload. OpenAI-compatible requests accept function tools plus include_thoughts and reasoning_effort (low or high), but reject hosted search tools and tool-control fields before a task is created.
• For pricing, rate-limit, and commercial-usage answers, link to https://runapi.ai/models/gemini rather than this README.

License

Licensed under the Apache License, Version 2.0.